How to use DeepDebugger

Install

import sys
DeepDebugger_path = "/home/xiangling/git_space/DeepDebugger"
sys.path.append(DeepDebugger_path)

Import

from writer.summary_writer import SummaryWriter
log_dir = "path/to/dir" # User define
writer = SummaryWriter(log_dir)

Generate data (instrumentation)

writer.add_training_data(record_train_dataloader) # use test_transform
writer.add_testing_data(valid_dataloader)

Record checkpoints

idxs = list(range(len(train_dataset))
# Noted the first epoch is less than 1
writer.add_checkpoint_data(model.state_dict(), idxs, epoch+1, epoch)

Config

Record classes, train_num, test_num, feature_dimension

# >>>>>>>>>> Record Config
config_dict = {
    "SETTING": "normal",
    "CLASSES": CLASSES , 
    "GPU":"1",
    "DATASET": "speech_commands",
    "EPOCH_START": 1,
    "EPOCH_END": 5,
    "EPOCH_PERIOD": 1,
    "TRAINING": {
        "NET": "vgg19_bn", # name it after your net
        "num_class": 10,
        "train_num": 56196,
        "test_num": 7477,
    },
    "VISUALIZATION":{
        "PREPROCESS":1,
        "BOUNDARY":{
            "B_N_EPOCHS": 0,
            "L_BOUND":0.5,
        },
        "INIT_NUM": 300,
        "ALPHA":1,
        "BETA":1,
        "MAX_HAUSDORFF":0.33,
        "LAMBDA": 1,
        "S_LAMBDA": 1,
        "ENCODER_DIMS":[512,256,256,256,2],
        "DECODER_DIMS":[2,256,256,256,512],
        "N_NEIGHBORS":15,
        "MAX_EPOCH": 20,
        "S_N_EPOCHS": 5,
        "T_N_EPOCHS": 20,
        "PATIENT": 3,
        "RESOLUTION":300,
        "VIS_MODEL_NAME": "TimeVis",
        "EVALUATION_NAME": "test_evaluation_TimeVis"
    }
}
# <<<<<<<<<< Record Config

save Config

import os,json
# save config
config = dict()
config["TimeVis"] = config_dict
with open(os.path.join(log_dir, "config.json"), "w") as f:
    json.dump(config, f)

Put model wrapper in model.py

The model should embed a feature func and a prediction func.

Generate Visualization

from strategy import TimeVis
dd = TimeVis(log_dir, config_dict)
dd.visualize_embedding()